Sanmina's AI Ramp Is Real, but It's a Story of Timing — Engineering Services Made Q3, Legacy Drags Q4

Sanmina's fiscal Q3 was a textbook beat-and-raise, with engineering services and a favorable mix lifting non-GAAP operating margin to 8% and EPS to $3.31 — up 116% year-over-year. But the more important news is what lies ahead: management explicitly guided Q4 ZT Systems revenue lower due to legacy program timing, and the next-generation accelerated compute program won't begin contributing until Q1 FY27. The market sold off from the June peak (the stock is down ~33% from its high), suggesting investors are pricing in the transition risk.

The Margin Surprise Is Real, but Partly One-Time

The beat was driven by more than just demand. As CFO Jonathan Faust explained, the contribution came from “new product launch services, or what we commonly refer to as non-recurring engineering services” “our non-GAAP operating margin of 8% and our non-GAAP diluted earnings per share of $3.31 both exceeded our outlook driven by strong execution, mix, and disciplined cost management, as well as contribution from new product launch services” — Jonathan Faust, CFO or Finance Executive · 2026-07-27. This is essentially pre-production work for the AMD-based accelerated compute rack program — labor-heavy, low-revenue work that inflates margins in the short term but isn't the long-run earnings engine. In Q&A, Faust reiterated that the ZT Systems business margin profile should settle in the 6–7% range once the ramp normalizes:

That is a great thing for us because it basically means that we are doing a lot of almost all of the pre-production work for the new accelerated compute program. The way that works... there is not a whole lot of revenue associated with it but, a lot of labor charges.

ZT Systems: A Tale of Two Programs

The quarter’s revenue mix underscores the transition. Core Sanmina grew 17% (ex-ZT), while ZT contributed $1.1B at the midpoint. But Q4 guidance for ZT is $0.8–1.0B, down from the prior implied run rate. The culprit: legacy programs that are phasing out. “This ZT Systems range is lower than the implied guidance provided last quarter and is driven completely by a few legacy programs, which, as I have mentioned before, can vary from quarter to quarter based on customer timing needs.” “This ZT Systems range is lower than the implied guidance provided last quarter and is driven completely by a few legacy programs, which, as I have mentioned before, can vary from quarter to quarter based on customer timing needs.” — Jonathan Faust, CFO or Finance Executive · 2026-07-27 Meanwhile, the new accelerated compute wins — including the previously announced Cerebras engagement — are on track but won't show up on the income statement until FY27. Management is confident in the $16B-plus revenue target for FY27, but the path is lumpy.

Working Capital: The Price of Growth

The company is deliberately investing ahead of the ramp. Inventory is already up 87% year-over-year (driven by ZT), and cash flow from operations was $124.5M in Q3, but free cash flow was just $23.6M after $100.9M of capex. CFO Faust noted: “As we have previously communicated, our long-term target net leverage range is 1.0x to 2.0x. We still expect our leverage to increase into our long-term range over time as we invest in working capital to support the growth of our business.” — Jonathan Faust, CFO or Finance Executive · 2026-07-27 This is a deliberate cash burn to build capacity for the AI infrastructure build-out — a bet that the AI system revenue will materialize at scale. The company's pretax ROIC was 39.1% in the quarter, well above its cost of capital, but that figure will likely compress as the balance sheet grows. Looking at the fundamentals, revenue has inflected sharply: Total Revenue reached $4.0B in the period ended April 2026, up 102% year-over-year, though this reflects only the first full quarter including ZT Systems.

The story is consistent with the broader AI infrastructure theme we're seeing across the tape. In the recent earnings reporters, names like AI infrastructure (from APLD) and AI compute (from TSLA) echo the same demand signals. Sanmina is levered to that wave, but it's a classic second-derivative story: the beat was real, but the guidance exposes the lumpiness of program transitions. Investors who bought the peak June rally need to buckle in for a choppy 90 days.

From a prior-quarter perspective, the company has been telegraphing this transition for a while. In January, Jure Sola said: “It's a transition year, a lot of work, but we're ahead of schedule.” — Jure Sola, Chief Executive Officer · 2026-04-27 And in April, Jonathan Faust noted: “we're very excited... to do so much business in this quarter around accelerated compute, and it's new platform business, not the old legacy products that ZT had.” — Jonathan Faust, Executive Vice President and Chief Financial Officer · 2026-04-27 The company is executing on that plan, but the market has to digest the timing.

Bottom line: Sanmina's AI story is intact, but the margin spike is inflated by NRE, and the ZT revenue dip in Q4 is a reminder that hyperscale program timing is unpredictable. The real test comes in FY27 as the next-gen accelerated compute ramp begins. Watch the cash conversion cycle and the balance sheet as the company borrows to fund growth — that's the metric that will tell you if this is a value-accretive expansion or a cash trap.